[PATCH] D54848: [PM] Port LoadStoreVectorizer to the new pass manager.

Fedor Sergeev via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Nov 30 05:40:40 PST 2018


fedor.sergeev added inline comments.


================
Comment at: test/Transforms/LoadStoreVectorizer/X86/correct-order.ll:2
 ; RUN: opt -mtriple=x86_64-unknown-linux-gnu -load-store-vectorizer -S -o - %s | FileCheck %s
+; RUN: opt -mtriple=x86_64-unknown-linux-gnu -aa-pipeline=basic-aa -passes='function(load-store-vectorizer)' -S -o - %s | FileCheck %s
 
----------------
markus wrote:
> fedor.sergeev wrote:
> > markus wrote:
> > > Unclear to my why -aa-pipeline=basic-aa is needed with the new PM (otherwise if not used AAResult contains no AAs and the test fails) while with the legacy PM -basicaa was not needed. Actually not clear to me if the two options are to be considered equivalent.
> > BasicAA is always available for function passes in legacy PM (see AAResultsWrapperPass::runOnFunction).
> > There is no unconditional AA analyses inserted into AA pipeline in new PM.
> In that case maybe we should have a -aa-pipeline=basic-aa option in all the .ll tests (I see that there are a few missing)? Not that it seems to matter for the tests passing but the pass has proven do depend on it for a few of them so might make sense.
I'm not sure about this.
If we need it all the time then perhaps we should implement a pipeline-by-default, similar to legacy pm?
I'll try to ping @chandlerc on that.


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D54848/new/

https://reviews.llvm.org/D54848





More information about the llvm-commits mailing list